正则化DropPath/drop_path用法示例(Python实现)

DropPath/drop_path 是一种正则化手段，其效果是将深度学习模型中的多分支结构随机”删除“，python中实现如下所示：

def drop_path(x, drop_prob: float = 0., training: bool = False):
    if drop_prob == 0. or not training:
        return x
    keep_prob = 1 - drop_prob
    shape = (x.shape[0],) + (1,) * (x.ndim - 1)
    random_tensor = keep_prob + torch.rand(shape, dtype=x.dtype, device=x.device)
    random_tensor.floor_()  # binarize
    output = x.div(keep_prob) * random_tensor
    return output

class DropPath(nn.Module):
    def __init__(self, drop_prob=None):
        super(DropPath, self).__init__()
        self.drop_prob = drop_prob

    def forward(self, x):
        return drop_path(x, self.drop_prob, self.training)

调用如下：

self.drop_path = DropPath(drop_prob) if drop_prob > 0. else nn.Identity()

x = x + self.drop_path(self.token_mixer(self.norm1(x)))
x = x + self.drop_path(self.mlp(self.norm2(x)))

看起来似乎有点迷茫，这怎么就随机删除了分支呢

实验如下：

import torch

drop_prob = 0.2
keep_prob = 1 - drop_prob
x = torch.randn(4, 3, 2, 2)
shape = (x.shape[0],) + (1,) * (x.ndim - 1)
random_tensor = keep_prob + torch.rand(shape, dtype=x.dtype, device=x.device)
random_tensor.floor_()
output = x.div(keep_prob) * random_tensor

输出：

x.size():[4,3,2,2]
x：
tensor([[[[ 1.3833, -0.3703],
          [-0.4608,  0.6955]],
         [[ 0.8306,  0.6882],
          [ 2.2375,  1.6158]],
         [[-0.7108,  1.0498],
          [ 0.6783,  1.5673]]],

[[[-0.0258, -1.7539],
          [-2.0789, -0.9648]],
         [[ 0.8598,  0.9351],
          [-0.3405,  0.0070]],
         [[ 0.3069, -1.5878],
          [-1.1333, -0.5932]]],

[[[ 1.0379,  0.6277],
          [ 0.0153, -0.4764]],
         [[ 1.0115, -0.0271],
          [ 1.6610, -0.2410]],
         [[ 0.0681, -2.0821],
          [ 0.6137,  0.1157]]],

[[[ 0.5350, -2.8424],
          [ 0.6648, -1.6652]],
         [[ 0.0122,  0.3389],
          [-1.1071, -0.6179]],
         [[-0.1843, -1.3026],
          [-0.3247,  0.3710]]]])

random_tensor.size():[4, 1, 1, 1]
random_tensor：
tensor([[[[0.]]],
        [[[1.]]],
        [[[1.]]],
        [[[1.]]]])

output.size():[4,3,2,2]
output：
tensor([[[[ 0.0000, -0.0000],
          [-0.0000,  0.0000]],
         [[ 0.0000,  0.0000],
          [ 0.0000,  0.0000]],
         [[-0.0000,  0.0000],
          [ 0.0000,  0.0000]]],

        [[[-0.0322, -2.1924],
          [-2.5986, -1.2060]],
         [[ 1.0748,  1.1689],
          [-0.4256,  0.0088]],
         [[ 0.3836, -1.9848],
          [-1.4166, -0.7415]]],

        [[[ 1.2974,  0.7846],
          [ 0.0192, -0.5955]],
         [[ 1.2644, -0.0339],
          [ 2.0762, -0.3012]],
         [[ 0.0851, -2.6027],
          [ 0.7671,  0.1446]]],

        [[[ 0.6687, -3.5530],
          [ 0.8310, -2.0815]],
         [[ 0.0152,  0.4236],
          [-1.3839, -0.7723]],
         [[-0.2303, -1.6282],
          [-0.4059,  0.4638]]]])

random_tensor作为是否保留分支的直接置0项，若drop_path的概率设为0.2，random_tensor中的数有0.2的概率为0，而output中被保留概率为0.8。

结合drop_path的调用，若x为输入的张量，其通道为[B,C,H,W]，那么drop_path的含义为在一个Batch_size中，随机有drop_prob的样本，不经过主干，而直接由分支进行恒等映射。

总结

到此这篇关于正则化DropPath/drop_path用法(Python实现)的文章就介绍到这了,更多相关正则化DropPath/drop_path内容请搜索我们以前的文章或继续浏览下面的相关文章希望大家以后多多支持我们！